xref: /openbmc/linux/arch/ia64/kernel/setup.c (revision b04b4f78)
1 /*
2  * Architecture-specific setup.
3  *
4  * Copyright (C) 1998-2001, 2003-2004 Hewlett-Packard Co
5  *	David Mosberger-Tang <davidm@hpl.hp.com>
6  *	Stephane Eranian <eranian@hpl.hp.com>
7  * Copyright (C) 2000, 2004 Intel Corp
8  * 	Rohit Seth <rohit.seth@intel.com>
9  * 	Suresh Siddha <suresh.b.siddha@intel.com>
10  * 	Gordon Jin <gordon.jin@intel.com>
11  * Copyright (C) 1999 VA Linux Systems
12  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
13  *
14  * 12/26/04 S.Siddha, G.Jin, R.Seth
15  *			Add multi-threading and multi-core detection
16  * 11/12/01 D.Mosberger Convert get_cpuinfo() to seq_file based show_cpuinfo().
17  * 04/04/00 D.Mosberger renamed cpu_initialized to cpu_online_map
18  * 03/31/00 R.Seth	cpu_initialized and current->processor fixes
19  * 02/04/00 D.Mosberger	some more get_cpuinfo fixes...
20  * 02/01/00 R.Seth	fixed get_cpuinfo for SMP
21  * 01/07/99 S.Eranian	added the support for command line argument
22  * 06/24/99 W.Drummond	added boot_cpu_data.
23  * 05/28/05 Z. Menyhart	Dynamic stride size for "flush_icache_range()"
24  */
25 #include <linux/module.h>
26 #include <linux/init.h>
27 
28 #include <linux/acpi.h>
29 #include <linux/bootmem.h>
30 #include <linux/console.h>
31 #include <linux/delay.h>
32 #include <linux/kernel.h>
33 #include <linux/reboot.h>
34 #include <linux/sched.h>
35 #include <linux/seq_file.h>
36 #include <linux/string.h>
37 #include <linux/threads.h>
38 #include <linux/screen_info.h>
39 #include <linux/dmi.h>
40 #include <linux/serial.h>
41 #include <linux/serial_core.h>
42 #include <linux/efi.h>
43 #include <linux/initrd.h>
44 #include <linux/pm.h>
45 #include <linux/cpufreq.h>
46 #include <linux/kexec.h>
47 #include <linux/crash_dump.h>
48 
49 #include <asm/ia32.h>
50 #include <asm/machvec.h>
51 #include <asm/mca.h>
52 #include <asm/meminit.h>
53 #include <asm/page.h>
54 #include <asm/paravirt.h>
55 #include <asm/paravirt_patch.h>
56 #include <asm/patch.h>
57 #include <asm/pgtable.h>
58 #include <asm/processor.h>
59 #include <asm/sal.h>
60 #include <asm/sections.h>
61 #include <asm/setup.h>
62 #include <asm/smp.h>
63 #include <asm/system.h>
64 #include <asm/tlbflush.h>
65 #include <asm/unistd.h>
66 #include <asm/hpsim.h>
67 
68 #if defined(CONFIG_SMP) && (IA64_CPU_SIZE > PAGE_SIZE)
69 # error "struct cpuinfo_ia64 too big!"
70 #endif
71 
72 #ifdef CONFIG_SMP
73 unsigned long __per_cpu_offset[NR_CPUS];
74 EXPORT_SYMBOL(__per_cpu_offset);
75 #endif
76 
77 DEFINE_PER_CPU(struct cpuinfo_ia64, cpu_info);
78 DEFINE_PER_CPU(unsigned long, local_per_cpu_offset);
79 unsigned long ia64_cycles_per_usec;
80 struct ia64_boot_param *ia64_boot_param;
81 struct screen_info screen_info;
82 unsigned long vga_console_iobase;
83 unsigned long vga_console_membase;
84 
85 static struct resource data_resource = {
86 	.name	= "Kernel data",
87 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
88 };
89 
90 static struct resource code_resource = {
91 	.name	= "Kernel code",
92 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
93 };
94 
95 static struct resource bss_resource = {
96 	.name	= "Kernel bss",
97 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
98 };
99 
100 unsigned long ia64_max_cacheline_size;
101 
102 int dma_get_cache_alignment(void)
103 {
104         return ia64_max_cacheline_size;
105 }
106 EXPORT_SYMBOL(dma_get_cache_alignment);
107 
108 unsigned long ia64_iobase;	/* virtual address for I/O accesses */
109 EXPORT_SYMBOL(ia64_iobase);
110 struct io_space io_space[MAX_IO_SPACES];
111 EXPORT_SYMBOL(io_space);
112 unsigned int num_io_spaces;
113 
114 /*
115  * "flush_icache_range()" needs to know what processor dependent stride size to use
116  * when it makes i-cache(s) coherent with d-caches.
117  */
118 #define	I_CACHE_STRIDE_SHIFT	5	/* Safest way to go: 32 bytes by 32 bytes */
119 unsigned long ia64_i_cache_stride_shift = ~0;
120 /*
121  * "clflush_cache_range()" needs to know what processor dependent stride size to
122  * use when it flushes cache lines including both d-cache and i-cache.
123  */
124 /* Safest way to go: 32 bytes by 32 bytes */
125 #define	CACHE_STRIDE_SHIFT	5
126 unsigned long ia64_cache_stride_shift = ~0;
127 
128 /*
129  * The merge_mask variable needs to be set to (max(iommu_page_size(iommu)) - 1).  This
130  * mask specifies a mask of address bits that must be 0 in order for two buffers to be
131  * mergeable by the I/O MMU (i.e., the end address of the first buffer and the start
132  * address of the second buffer must be aligned to (merge_mask+1) in order to be
133  * mergeable).  By default, we assume there is no I/O MMU which can merge physically
134  * discontiguous buffers, so we set the merge_mask to ~0UL, which corresponds to a iommu
135  * page-size of 2^64.
136  */
137 unsigned long ia64_max_iommu_merge_mask = ~0UL;
138 EXPORT_SYMBOL(ia64_max_iommu_merge_mask);
139 
140 /*
141  * We use a special marker for the end of memory and it uses the extra (+1) slot
142  */
143 struct rsvd_region rsvd_region[IA64_MAX_RSVD_REGIONS + 1] __initdata;
144 int num_rsvd_regions __initdata;
145 
146 
147 /*
148  * Filter incoming memory segments based on the primitive map created from the boot
149  * parameters. Segments contained in the map are removed from the memory ranges. A
150  * caller-specified function is called with the memory ranges that remain after filtering.
151  * This routine does not assume the incoming segments are sorted.
152  */
153 int __init
154 filter_rsvd_memory (unsigned long start, unsigned long end, void *arg)
155 {
156 	unsigned long range_start, range_end, prev_start;
157 	void (*func)(unsigned long, unsigned long, int);
158 	int i;
159 
160 #if IGNORE_PFN0
161 	if (start == PAGE_OFFSET) {
162 		printk(KERN_WARNING "warning: skipping physical page 0\n");
163 		start += PAGE_SIZE;
164 		if (start >= end) return 0;
165 	}
166 #endif
167 	/*
168 	 * lowest possible address(walker uses virtual)
169 	 */
170 	prev_start = PAGE_OFFSET;
171 	func = arg;
172 
173 	for (i = 0; i < num_rsvd_regions; ++i) {
174 		range_start = max(start, prev_start);
175 		range_end   = min(end, rsvd_region[i].start);
176 
177 		if (range_start < range_end)
178 			call_pernode_memory(__pa(range_start), range_end - range_start, func);
179 
180 		/* nothing more available in this segment */
181 		if (range_end == end) return 0;
182 
183 		prev_start = rsvd_region[i].end;
184 	}
185 	/* end of memory marker allows full processing inside loop body */
186 	return 0;
187 }
188 
189 /*
190  * Similar to "filter_rsvd_memory()", but the reserved memory ranges
191  * are not filtered out.
192  */
193 int __init
194 filter_memory(unsigned long start, unsigned long end, void *arg)
195 {
196 	void (*func)(unsigned long, unsigned long, int);
197 
198 #if IGNORE_PFN0
199 	if (start == PAGE_OFFSET) {
200 		printk(KERN_WARNING "warning: skipping physical page 0\n");
201 		start += PAGE_SIZE;
202 		if (start >= end)
203 			return 0;
204 	}
205 #endif
206 	func = arg;
207 	if (start < end)
208 		call_pernode_memory(__pa(start), end - start, func);
209 	return 0;
210 }
211 
212 static void __init
213 sort_regions (struct rsvd_region *rsvd_region, int max)
214 {
215 	int j;
216 
217 	/* simple bubble sorting */
218 	while (max--) {
219 		for (j = 0; j < max; ++j) {
220 			if (rsvd_region[j].start > rsvd_region[j+1].start) {
221 				struct rsvd_region tmp;
222 				tmp = rsvd_region[j];
223 				rsvd_region[j] = rsvd_region[j + 1];
224 				rsvd_region[j + 1] = tmp;
225 			}
226 		}
227 	}
228 }
229 
230 /*
231  * Request address space for all standard resources
232  */
233 static int __init register_memory(void)
234 {
235 	code_resource.start = ia64_tpa(_text);
236 	code_resource.end   = ia64_tpa(_etext) - 1;
237 	data_resource.start = ia64_tpa(_etext);
238 	data_resource.end   = ia64_tpa(_edata) - 1;
239 	bss_resource.start  = ia64_tpa(__bss_start);
240 	bss_resource.end    = ia64_tpa(_end) - 1;
241 	efi_initialize_iomem_resources(&code_resource, &data_resource,
242 			&bss_resource);
243 
244 	return 0;
245 }
246 
247 __initcall(register_memory);
248 
249 
250 #ifdef CONFIG_KEXEC
251 
252 /*
253  * This function checks if the reserved crashkernel is allowed on the specific
254  * IA64 machine flavour. Machines without an IO TLB use swiotlb and require
255  * some memory below 4 GB (i.e. in 32 bit area), see the implementation of
256  * lib/swiotlb.c. The hpzx1 architecture has an IO TLB but cannot use that
257  * in kdump case. See the comment in sba_init() in sba_iommu.c.
258  *
259  * So, the only machvec that really supports loading the kdump kernel
260  * over 4 GB is "sn2".
261  */
262 static int __init check_crashkernel_memory(unsigned long pbase, size_t size)
263 {
264 	if (ia64_platform_is("sn2") || ia64_platform_is("uv"))
265 		return 1;
266 	else
267 		return pbase < (1UL << 32);
268 }
269 
270 static void __init setup_crashkernel(unsigned long total, int *n)
271 {
272 	unsigned long long base = 0, size = 0;
273 	int ret;
274 
275 	ret = parse_crashkernel(boot_command_line, total,
276 			&size, &base);
277 	if (ret == 0 && size > 0) {
278 		if (!base) {
279 			sort_regions(rsvd_region, *n);
280 			base = kdump_find_rsvd_region(size,
281 					rsvd_region, *n);
282 		}
283 
284 		if (!check_crashkernel_memory(base, size)) {
285 			pr_warning("crashkernel: There would be kdump memory "
286 				"at %ld GB but this is unusable because it "
287 				"must\nbe below 4 GB. Change the memory "
288 				"configuration of the machine.\n",
289 				(unsigned long)(base >> 30));
290 			return;
291 		}
292 
293 		if (base != ~0UL) {
294 			printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
295 					"for crashkernel (System RAM: %ldMB)\n",
296 					(unsigned long)(size >> 20),
297 					(unsigned long)(base >> 20),
298 					(unsigned long)(total >> 20));
299 			rsvd_region[*n].start =
300 				(unsigned long)__va(base);
301 			rsvd_region[*n].end =
302 				(unsigned long)__va(base + size);
303 			(*n)++;
304 			crashk_res.start = base;
305 			crashk_res.end = base + size - 1;
306 		}
307 	}
308 	efi_memmap_res.start = ia64_boot_param->efi_memmap;
309 	efi_memmap_res.end = efi_memmap_res.start +
310 		ia64_boot_param->efi_memmap_size;
311 	boot_param_res.start = __pa(ia64_boot_param);
312 	boot_param_res.end = boot_param_res.start +
313 		sizeof(*ia64_boot_param);
314 }
315 #else
316 static inline void __init setup_crashkernel(unsigned long total, int *n)
317 {}
318 #endif
319 
320 /**
321  * reserve_memory - setup reserved memory areas
322  *
323  * Setup the reserved memory areas set aside for the boot parameters,
324  * initrd, etc.  There are currently %IA64_MAX_RSVD_REGIONS defined,
325  * see arch/ia64/include/asm/meminit.h if you need to define more.
326  */
327 void __init
328 reserve_memory (void)
329 {
330 	int n = 0;
331 	unsigned long total_memory;
332 
333 	/*
334 	 * none of the entries in this table overlap
335 	 */
336 	rsvd_region[n].start = (unsigned long) ia64_boot_param;
337 	rsvd_region[n].end   = rsvd_region[n].start + sizeof(*ia64_boot_param);
338 	n++;
339 
340 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->efi_memmap);
341 	rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->efi_memmap_size;
342 	n++;
343 
344 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->command_line);
345 	rsvd_region[n].end   = (rsvd_region[n].start
346 				+ strlen(__va(ia64_boot_param->command_line)) + 1);
347 	n++;
348 
349 	rsvd_region[n].start = (unsigned long) ia64_imva((void *)KERNEL_START);
350 	rsvd_region[n].end   = (unsigned long) ia64_imva(_end);
351 	n++;
352 
353 	n += paravirt_reserve_memory(&rsvd_region[n]);
354 
355 #ifdef CONFIG_BLK_DEV_INITRD
356 	if (ia64_boot_param->initrd_start) {
357 		rsvd_region[n].start = (unsigned long)__va(ia64_boot_param->initrd_start);
358 		rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->initrd_size;
359 		n++;
360 	}
361 #endif
362 
363 #ifdef CONFIG_CRASH_DUMP
364 	if (reserve_elfcorehdr(&rsvd_region[n].start,
365 			       &rsvd_region[n].end) == 0)
366 		n++;
367 #endif
368 
369 	total_memory = efi_memmap_init(&rsvd_region[n].start, &rsvd_region[n].end);
370 	n++;
371 
372 	setup_crashkernel(total_memory, &n);
373 
374 	/* end of memory marker */
375 	rsvd_region[n].start = ~0UL;
376 	rsvd_region[n].end   = ~0UL;
377 	n++;
378 
379 	num_rsvd_regions = n;
380 	BUG_ON(IA64_MAX_RSVD_REGIONS + 1 < n);
381 
382 	sort_regions(rsvd_region, num_rsvd_regions);
383 }
384 
385 
386 /**
387  * find_initrd - get initrd parameters from the boot parameter structure
388  *
389  * Grab the initrd start and end from the boot parameter struct given us by
390  * the boot loader.
391  */
392 void __init
393 find_initrd (void)
394 {
395 #ifdef CONFIG_BLK_DEV_INITRD
396 	if (ia64_boot_param->initrd_start) {
397 		initrd_start = (unsigned long)__va(ia64_boot_param->initrd_start);
398 		initrd_end   = initrd_start+ia64_boot_param->initrd_size;
399 
400 		printk(KERN_INFO "Initial ramdisk at: 0x%lx (%lu bytes)\n",
401 		       initrd_start, ia64_boot_param->initrd_size);
402 	}
403 #endif
404 }
405 
406 static void __init
407 io_port_init (void)
408 {
409 	unsigned long phys_iobase;
410 
411 	/*
412 	 * Set `iobase' based on the EFI memory map or, failing that, the
413 	 * value firmware left in ar.k0.
414 	 *
415 	 * Note that in ia32 mode, IN/OUT instructions use ar.k0 to compute
416 	 * the port's virtual address, so ia32_load_state() loads it with a
417 	 * user virtual address.  But in ia64 mode, glibc uses the
418 	 * *physical* address in ar.k0 to mmap the appropriate area from
419 	 * /dev/mem, and the inX()/outX() interfaces use MMIO.  In both
420 	 * cases, user-mode can only use the legacy 0-64K I/O port space.
421 	 *
422 	 * ar.k0 is not involved in kernel I/O port accesses, which can use
423 	 * any of the I/O port spaces and are done via MMIO using the
424 	 * virtual mmio_base from the appropriate io_space[].
425 	 */
426 	phys_iobase = efi_get_iobase();
427 	if (!phys_iobase) {
428 		phys_iobase = ia64_get_kr(IA64_KR_IO_BASE);
429 		printk(KERN_INFO "No I/O port range found in EFI memory map, "
430 			"falling back to AR.KR0 (0x%lx)\n", phys_iobase);
431 	}
432 	ia64_iobase = (unsigned long) ioremap(phys_iobase, 0);
433 	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
434 
435 	/* setup legacy IO port space */
436 	io_space[0].mmio_base = ia64_iobase;
437 	io_space[0].sparse = 1;
438 	num_io_spaces = 1;
439 }
440 
441 /**
442  * early_console_setup - setup debugging console
443  *
444  * Consoles started here require little enough setup that we can start using
445  * them very early in the boot process, either right after the machine
446  * vector initialization, or even before if the drivers can detect their hw.
447  *
448  * Returns non-zero if a console couldn't be setup.
449  */
450 static inline int __init
451 early_console_setup (char *cmdline)
452 {
453 	int earlycons = 0;
454 
455 #ifdef CONFIG_SERIAL_SGI_L1_CONSOLE
456 	{
457 		extern int sn_serial_console_early_setup(void);
458 		if (!sn_serial_console_early_setup())
459 			earlycons++;
460 	}
461 #endif
462 #ifdef CONFIG_EFI_PCDP
463 	if (!efi_setup_pcdp_console(cmdline))
464 		earlycons++;
465 #endif
466 	if (!simcons_register())
467 		earlycons++;
468 
469 	return (earlycons) ? 0 : -1;
470 }
471 
472 static inline void
473 mark_bsp_online (void)
474 {
475 #ifdef CONFIG_SMP
476 	/* If we register an early console, allow CPU 0 to printk */
477 	cpu_set(smp_processor_id(), cpu_online_map);
478 #endif
479 }
480 
481 static __initdata int nomca;
482 static __init int setup_nomca(char *s)
483 {
484 	nomca = 1;
485 	return 0;
486 }
487 early_param("nomca", setup_nomca);
488 
489 /*
490  * Note: elfcorehdr_addr is not just limited to vmcore. It is also used by
491  * is_kdump_kernel() to determine if we are booting after a panic. Hence
492  * ifdef it under CONFIG_CRASH_DUMP and not CONFIG_PROC_VMCORE.
493  */
494 #ifdef CONFIG_CRASH_DUMP
495 /* elfcorehdr= specifies the location of elf core header
496  * stored by the crashed kernel.
497  */
498 static int __init parse_elfcorehdr(char *arg)
499 {
500 	if (!arg)
501 		return -EINVAL;
502 
503         elfcorehdr_addr = memparse(arg, &arg);
504 	return 0;
505 }
506 early_param("elfcorehdr", parse_elfcorehdr);
507 
508 int __init reserve_elfcorehdr(unsigned long *start, unsigned long *end)
509 {
510 	unsigned long length;
511 
512 	/* We get the address using the kernel command line,
513 	 * but the size is extracted from the EFI tables.
514 	 * Both address and size are required for reservation
515 	 * to work properly.
516 	 */
517 
518 	if (!is_vmcore_usable())
519 		return -EINVAL;
520 
521 	if ((length = vmcore_find_descriptor_size(elfcorehdr_addr)) == 0) {
522 		vmcore_unusable();
523 		return -EINVAL;
524 	}
525 
526 	*start = (unsigned long)__va(elfcorehdr_addr);
527 	*end = *start + length;
528 	return 0;
529 }
530 
531 #endif /* CONFIG_PROC_VMCORE */
532 
533 void __init
534 setup_arch (char **cmdline_p)
535 {
536 	unw_init();
537 
538 	paravirt_arch_setup_early();
539 
540 	ia64_patch_vtop((u64) __start___vtop_patchlist, (u64) __end___vtop_patchlist);
541 	paravirt_patch_apply();
542 
543 	*cmdline_p = __va(ia64_boot_param->command_line);
544 	strlcpy(boot_command_line, *cmdline_p, COMMAND_LINE_SIZE);
545 
546 	efi_init();
547 	io_port_init();
548 
549 #ifdef CONFIG_IA64_GENERIC
550 	/* machvec needs to be parsed from the command line
551 	 * before parse_early_param() is called to ensure
552 	 * that ia64_mv is initialised before any command line
553 	 * settings may cause console setup to occur
554 	 */
555 	machvec_init_from_cmdline(*cmdline_p);
556 #endif
557 
558 	parse_early_param();
559 
560 	if (early_console_setup(*cmdline_p) == 0)
561 		mark_bsp_online();
562 
563 #ifdef CONFIG_ACPI
564 	/* Initialize the ACPI boot-time table parser */
565 	acpi_table_init();
566 	early_acpi_boot_init();
567 # ifdef CONFIG_ACPI_NUMA
568 	acpi_numa_init();
569 #ifdef CONFIG_ACPI_HOTPLUG_CPU
570 	prefill_possible_map();
571 #endif
572 	per_cpu_scan_finalize((cpus_weight(early_cpu_possible_map) == 0 ?
573 		32 : cpus_weight(early_cpu_possible_map)),
574 		additional_cpus > 0 ? additional_cpus : 0);
575 # endif
576 #else
577 # ifdef CONFIG_SMP
578 	smp_build_cpu_map();	/* happens, e.g., with the Ski simulator */
579 # endif
580 #endif /* CONFIG_APCI_BOOT */
581 
582 	find_memory();
583 
584 	/* process SAL system table: */
585 	ia64_sal_init(__va(efi.sal_systab));
586 
587 #ifdef CONFIG_ITANIUM
588 	ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
589 #else
590 	{
591 		u64 num_phys_stacked;
592 
593 		if (ia64_pal_rse_info(&num_phys_stacked, 0) == 0 && num_phys_stacked > 96)
594 			ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
595 	}
596 #endif
597 
598 #ifdef CONFIG_SMP
599 	cpu_physical_id(0) = hard_smp_processor_id();
600 #endif
601 
602 	cpu_init();	/* initialize the bootstrap CPU */
603 	mmu_context_init();	/* initialize context_id bitmap */
604 
605 #ifdef CONFIG_ACPI
606 	acpi_boot_init();
607 #endif
608 
609 	paravirt_banner();
610 	paravirt_arch_setup_console(cmdline_p);
611 
612 #ifdef CONFIG_VT
613 	if (!conswitchp) {
614 # if defined(CONFIG_DUMMY_CONSOLE)
615 		conswitchp = &dummy_con;
616 # endif
617 # if defined(CONFIG_VGA_CONSOLE)
618 		/*
619 		 * Non-legacy systems may route legacy VGA MMIO range to system
620 		 * memory.  vga_con probes the MMIO hole, so memory looks like
621 		 * a VGA device to it.  The EFI memory map can tell us if it's
622 		 * memory so we can avoid this problem.
623 		 */
624 		if (efi_mem_type(0xA0000) != EFI_CONVENTIONAL_MEMORY)
625 			conswitchp = &vga_con;
626 # endif
627 	}
628 #endif
629 
630 	/* enable IA-64 Machine Check Abort Handling unless disabled */
631 	if (paravirt_arch_setup_nomca())
632 		nomca = 1;
633 	if (!nomca)
634 		ia64_mca_init();
635 
636 	platform_setup(cmdline_p);
637 #ifndef CONFIG_IA64_HP_SIM
638 	check_sal_cache_flush();
639 #endif
640 	paging_init();
641 }
642 
643 /*
644  * Display cpu info for all CPUs.
645  */
646 static int
647 show_cpuinfo (struct seq_file *m, void *v)
648 {
649 #ifdef CONFIG_SMP
650 #	define lpj	c->loops_per_jiffy
651 #	define cpunum	c->cpu
652 #else
653 #	define lpj	loops_per_jiffy
654 #	define cpunum	0
655 #endif
656 	static struct {
657 		unsigned long mask;
658 		const char *feature_name;
659 	} feature_bits[] = {
660 		{ 1UL << 0, "branchlong" },
661 		{ 1UL << 1, "spontaneous deferral"},
662 		{ 1UL << 2, "16-byte atomic ops" }
663 	};
664 	char features[128], *cp, *sep;
665 	struct cpuinfo_ia64 *c = v;
666 	unsigned long mask;
667 	unsigned long proc_freq;
668 	int i, size;
669 
670 	mask = c->features;
671 
672 	/* build the feature string: */
673 	memcpy(features, "standard", 9);
674 	cp = features;
675 	size = sizeof(features);
676 	sep = "";
677 	for (i = 0; i < ARRAY_SIZE(feature_bits) && size > 1; ++i) {
678 		if (mask & feature_bits[i].mask) {
679 			cp += snprintf(cp, size, "%s%s", sep,
680 				       feature_bits[i].feature_name),
681 			sep = ", ";
682 			mask &= ~feature_bits[i].mask;
683 			size = sizeof(features) - (cp - features);
684 		}
685 	}
686 	if (mask && size > 1) {
687 		/* print unknown features as a hex value */
688 		snprintf(cp, size, "%s0x%lx", sep, mask);
689 	}
690 
691 	proc_freq = cpufreq_quick_get(cpunum);
692 	if (!proc_freq)
693 		proc_freq = c->proc_freq / 1000;
694 
695 	seq_printf(m,
696 		   "processor  : %d\n"
697 		   "vendor     : %s\n"
698 		   "arch       : IA-64\n"
699 		   "family     : %u\n"
700 		   "model      : %u\n"
701 		   "model name : %s\n"
702 		   "revision   : %u\n"
703 		   "archrev    : %u\n"
704 		   "features   : %s\n"
705 		   "cpu number : %lu\n"
706 		   "cpu regs   : %u\n"
707 		   "cpu MHz    : %lu.%03lu\n"
708 		   "itc MHz    : %lu.%06lu\n"
709 		   "BogoMIPS   : %lu.%02lu\n",
710 		   cpunum, c->vendor, c->family, c->model,
711 		   c->model_name, c->revision, c->archrev,
712 		   features, c->ppn, c->number,
713 		   proc_freq / 1000, proc_freq % 1000,
714 		   c->itc_freq / 1000000, c->itc_freq % 1000000,
715 		   lpj*HZ/500000, (lpj*HZ/5000) % 100);
716 #ifdef CONFIG_SMP
717 	seq_printf(m, "siblings   : %u\n", cpus_weight(cpu_core_map[cpunum]));
718 	if (c->socket_id != -1)
719 		seq_printf(m, "physical id: %u\n", c->socket_id);
720 	if (c->threads_per_core > 1 || c->cores_per_socket > 1)
721 		seq_printf(m,
722 			   "core id    : %u\n"
723 			   "thread id  : %u\n",
724 			   c->core_id, c->thread_id);
725 #endif
726 	seq_printf(m,"\n");
727 
728 	return 0;
729 }
730 
731 static void *
732 c_start (struct seq_file *m, loff_t *pos)
733 {
734 #ifdef CONFIG_SMP
735 	while (*pos < nr_cpu_ids && !cpu_online(*pos))
736 		++*pos;
737 #endif
738 	return *pos < nr_cpu_ids ? cpu_data(*pos) : NULL;
739 }
740 
741 static void *
742 c_next (struct seq_file *m, void *v, loff_t *pos)
743 {
744 	++*pos;
745 	return c_start(m, pos);
746 }
747 
748 static void
749 c_stop (struct seq_file *m, void *v)
750 {
751 }
752 
753 const struct seq_operations cpuinfo_op = {
754 	.start =	c_start,
755 	.next =		c_next,
756 	.stop =		c_stop,
757 	.show =		show_cpuinfo
758 };
759 
760 #define MAX_BRANDS	8
761 static char brandname[MAX_BRANDS][128];
762 
763 static char * __cpuinit
764 get_model_name(__u8 family, __u8 model)
765 {
766 	static int overflow;
767 	char brand[128];
768 	int i;
769 
770 	memcpy(brand, "Unknown", 8);
771 	if (ia64_pal_get_brand_info(brand)) {
772 		if (family == 0x7)
773 			memcpy(brand, "Merced", 7);
774 		else if (family == 0x1f) switch (model) {
775 			case 0: memcpy(brand, "McKinley", 9); break;
776 			case 1: memcpy(brand, "Madison", 8); break;
777 			case 2: memcpy(brand, "Madison up to 9M cache", 23); break;
778 		}
779 	}
780 	for (i = 0; i < MAX_BRANDS; i++)
781 		if (strcmp(brandname[i], brand) == 0)
782 			return brandname[i];
783 	for (i = 0; i < MAX_BRANDS; i++)
784 		if (brandname[i][0] == '\0')
785 			return strcpy(brandname[i], brand);
786 	if (overflow++ == 0)
787 		printk(KERN_ERR
788 		       "%s: Table overflow. Some processor model information will be missing\n",
789 		       __func__);
790 	return "Unknown";
791 }
792 
793 static void __cpuinit
794 identify_cpu (struct cpuinfo_ia64 *c)
795 {
796 	union {
797 		unsigned long bits[5];
798 		struct {
799 			/* id 0 & 1: */
800 			char vendor[16];
801 
802 			/* id 2 */
803 			u64 ppn;		/* processor serial number */
804 
805 			/* id 3: */
806 			unsigned number		:  8;
807 			unsigned revision	:  8;
808 			unsigned model		:  8;
809 			unsigned family		:  8;
810 			unsigned archrev	:  8;
811 			unsigned reserved	: 24;
812 
813 			/* id 4: */
814 			u64 features;
815 		} field;
816 	} cpuid;
817 	pal_vm_info_1_u_t vm1;
818 	pal_vm_info_2_u_t vm2;
819 	pal_status_t status;
820 	unsigned long impl_va_msb = 50, phys_addr_size = 44;	/* Itanium defaults */
821 	int i;
822 	for (i = 0; i < 5; ++i)
823 		cpuid.bits[i] = ia64_get_cpuid(i);
824 
825 	memcpy(c->vendor, cpuid.field.vendor, 16);
826 #ifdef CONFIG_SMP
827 	c->cpu = smp_processor_id();
828 
829 	/* below default values will be overwritten  by identify_siblings()
830 	 * for Multi-Threading/Multi-Core capable CPUs
831 	 */
832 	c->threads_per_core = c->cores_per_socket = c->num_log = 1;
833 	c->socket_id = -1;
834 
835 	identify_siblings(c);
836 
837 	if (c->threads_per_core > smp_num_siblings)
838 		smp_num_siblings = c->threads_per_core;
839 #endif
840 	c->ppn = cpuid.field.ppn;
841 	c->number = cpuid.field.number;
842 	c->revision = cpuid.field.revision;
843 	c->model = cpuid.field.model;
844 	c->family = cpuid.field.family;
845 	c->archrev = cpuid.field.archrev;
846 	c->features = cpuid.field.features;
847 	c->model_name = get_model_name(c->family, c->model);
848 
849 	status = ia64_pal_vm_summary(&vm1, &vm2);
850 	if (status == PAL_STATUS_SUCCESS) {
851 		impl_va_msb = vm2.pal_vm_info_2_s.impl_va_msb;
852 		phys_addr_size = vm1.pal_vm_info_1_s.phys_add_size;
853 	}
854 	c->unimpl_va_mask = ~((7L<<61) | ((1L << (impl_va_msb + 1)) - 1));
855 	c->unimpl_pa_mask = ~((1L<<63) | ((1L << phys_addr_size) - 1));
856 }
857 
858 void __init
859 setup_per_cpu_areas (void)
860 {
861 	/* start_kernel() requires this... */
862 }
863 
864 /*
865  * Do the following calculations:
866  *
867  * 1. the max. cache line size.
868  * 2. the minimum of the i-cache stride sizes for "flush_icache_range()".
869  * 3. the minimum of the cache stride sizes for "clflush_cache_range()".
870  */
871 static void __cpuinit
872 get_cache_info(void)
873 {
874 	unsigned long line_size, max = 1;
875 	u64 l, levels, unique_caches;
876         pal_cache_config_info_t cci;
877         s64 status;
878 
879         status = ia64_pal_cache_summary(&levels, &unique_caches);
880         if (status != 0) {
881                 printk(KERN_ERR "%s: ia64_pal_cache_summary() failed (status=%ld)\n",
882                        __func__, status);
883                 max = SMP_CACHE_BYTES;
884 		/* Safest setup for "flush_icache_range()" */
885 		ia64_i_cache_stride_shift = I_CACHE_STRIDE_SHIFT;
886 		/* Safest setup for "clflush_cache_range()" */
887 		ia64_cache_stride_shift = CACHE_STRIDE_SHIFT;
888 		goto out;
889         }
890 
891 	for (l = 0; l < levels; ++l) {
892 		/* cache_type (data_or_unified)=2 */
893 		status = ia64_pal_cache_config_info(l, 2, &cci);
894 		if (status != 0) {
895 			printk(KERN_ERR
896 			       "%s: ia64_pal_cache_config_info(l=%lu, 2) failed (status=%ld)\n",
897 			       __func__, l, status);
898 			max = SMP_CACHE_BYTES;
899 			/* The safest setup for "flush_icache_range()" */
900 			cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
901 			/* The safest setup for "clflush_cache_range()" */
902 			ia64_cache_stride_shift = CACHE_STRIDE_SHIFT;
903 			cci.pcci_unified = 1;
904 		} else {
905 			if (cci.pcci_stride < ia64_cache_stride_shift)
906 				ia64_cache_stride_shift = cci.pcci_stride;
907 
908 			line_size = 1 << cci.pcci_line_size;
909 			if (line_size > max)
910 				max = line_size;
911 		}
912 
913 		if (!cci.pcci_unified) {
914 			/* cache_type (instruction)=1*/
915 			status = ia64_pal_cache_config_info(l, 1, &cci);
916 			if (status != 0) {
917 				printk(KERN_ERR
918 				"%s: ia64_pal_cache_config_info(l=%lu, 1) failed (status=%ld)\n",
919 					__func__, l, status);
920 				/* The safest setup for "flush_icache_range()" */
921 				cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
922 			}
923 		}
924 		if (cci.pcci_stride < ia64_i_cache_stride_shift)
925 			ia64_i_cache_stride_shift = cci.pcci_stride;
926 	}
927   out:
928 	if (max > ia64_max_cacheline_size)
929 		ia64_max_cacheline_size = max;
930 }
931 
932 /*
933  * cpu_init() initializes state that is per-CPU.  This function acts
934  * as a 'CPU state barrier', nothing should get across.
935  */
936 void __cpuinit
937 cpu_init (void)
938 {
939 	extern void __cpuinit ia64_mmu_init (void *);
940 	static unsigned long max_num_phys_stacked = IA64_NUM_PHYS_STACK_REG;
941 	unsigned long num_phys_stacked;
942 	pal_vm_info_2_u_t vmi;
943 	unsigned int max_ctx;
944 	struct cpuinfo_ia64 *cpu_info;
945 	void *cpu_data;
946 
947 	cpu_data = per_cpu_init();
948 #ifdef CONFIG_SMP
949 	/*
950 	 * insert boot cpu into sibling and core mapes
951 	 * (must be done after per_cpu area is setup)
952 	 */
953 	if (smp_processor_id() == 0) {
954 		cpu_set(0, per_cpu(cpu_sibling_map, 0));
955 		cpu_set(0, cpu_core_map[0]);
956 	} else {
957 		/*
958 		 * Set ar.k3 so that assembly code in MCA handler can compute
959 		 * physical addresses of per cpu variables with a simple:
960 		 *   phys = ar.k3 + &per_cpu_var
961 		 * and the alt-dtlb-miss handler can set per-cpu mapping into
962 		 * the TLB when needed. head.S already did this for cpu0.
963 		 */
964 		ia64_set_kr(IA64_KR_PER_CPU_DATA,
965 			    ia64_tpa(cpu_data) - (long) __per_cpu_start);
966 	}
967 #endif
968 
969 	get_cache_info();
970 
971 	/*
972 	 * We can't pass "local_cpu_data" to identify_cpu() because we haven't called
973 	 * ia64_mmu_init() yet.  And we can't call ia64_mmu_init() first because it
974 	 * depends on the data returned by identify_cpu().  We break the dependency by
975 	 * accessing cpu_data() through the canonical per-CPU address.
976 	 */
977 	cpu_info = cpu_data + ((char *) &__ia64_per_cpu_var(cpu_info) - __per_cpu_start);
978 	identify_cpu(cpu_info);
979 
980 #ifdef CONFIG_MCKINLEY
981 	{
982 #		define FEATURE_SET 16
983 		struct ia64_pal_retval iprv;
984 
985 		if (cpu_info->family == 0x1f) {
986 			PAL_CALL_PHYS(iprv, PAL_PROC_GET_FEATURES, 0, FEATURE_SET, 0);
987 			if ((iprv.status == 0) && (iprv.v0 & 0x80) && (iprv.v2 & 0x80))
988 				PAL_CALL_PHYS(iprv, PAL_PROC_SET_FEATURES,
989 				              (iprv.v1 | 0x80), FEATURE_SET, 0);
990 		}
991 	}
992 #endif
993 
994 	/* Clear the stack memory reserved for pt_regs: */
995 	memset(task_pt_regs(current), 0, sizeof(struct pt_regs));
996 
997 	ia64_set_kr(IA64_KR_FPU_OWNER, 0);
998 
999 	/*
1000 	 * Initialize the page-table base register to a global
1001 	 * directory with all zeroes.  This ensure that we can handle
1002 	 * TLB-misses to user address-space even before we created the
1003 	 * first user address-space.  This may happen, e.g., due to
1004 	 * aggressive use of lfetch.fault.
1005 	 */
1006 	ia64_set_kr(IA64_KR_PT_BASE, __pa(ia64_imva(empty_zero_page)));
1007 
1008 	/*
1009 	 * Initialize default control register to defer speculative faults except
1010 	 * for those arising from TLB misses, which are not deferred.  The
1011 	 * kernel MUST NOT depend on a particular setting of these bits (in other words,
1012 	 * the kernel must have recovery code for all speculative accesses).  Turn on
1013 	 * dcr.lc as per recommendation by the architecture team.  Most IA-32 apps
1014 	 * shouldn't be affected by this (moral: keep your ia32 locks aligned and you'll
1015 	 * be fine).
1016 	 */
1017 	ia64_setreg(_IA64_REG_CR_DCR,  (  IA64_DCR_DP | IA64_DCR_DK | IA64_DCR_DX | IA64_DCR_DR
1018 					| IA64_DCR_DA | IA64_DCR_DD | IA64_DCR_LC));
1019 	atomic_inc(&init_mm.mm_count);
1020 	current->active_mm = &init_mm;
1021 	BUG_ON(current->mm);
1022 
1023 	ia64_mmu_init(ia64_imva(cpu_data));
1024 	ia64_mca_cpu_init(ia64_imva(cpu_data));
1025 
1026 #ifdef CONFIG_IA32_SUPPORT
1027 	ia32_cpu_init();
1028 #endif
1029 
1030 	/* Clear ITC to eliminate sched_clock() overflows in human time.  */
1031 	ia64_set_itc(0);
1032 
1033 	/* disable all local interrupt sources: */
1034 	ia64_set_itv(1 << 16);
1035 	ia64_set_lrr0(1 << 16);
1036 	ia64_set_lrr1(1 << 16);
1037 	ia64_setreg(_IA64_REG_CR_PMV, 1 << 16);
1038 	ia64_setreg(_IA64_REG_CR_CMCV, 1 << 16);
1039 
1040 	/* clear TPR & XTP to enable all interrupt classes: */
1041 	ia64_setreg(_IA64_REG_CR_TPR, 0);
1042 
1043 	/* Clear any pending interrupts left by SAL/EFI */
1044 	while (ia64_get_ivr() != IA64_SPURIOUS_INT_VECTOR)
1045 		ia64_eoi();
1046 
1047 #ifdef CONFIG_SMP
1048 	normal_xtp();
1049 #endif
1050 
1051 	/* set ia64_ctx.max_rid to the maximum RID that is supported by all CPUs: */
1052 	if (ia64_pal_vm_summary(NULL, &vmi) == 0) {
1053 		max_ctx = (1U << (vmi.pal_vm_info_2_s.rid_size - 3)) - 1;
1054 		setup_ptcg_sem(vmi.pal_vm_info_2_s.max_purges, NPTCG_FROM_PAL);
1055 	} else {
1056 		printk(KERN_WARNING "cpu_init: PAL VM summary failed, assuming 18 RID bits\n");
1057 		max_ctx = (1U << 15) - 1;	/* use architected minimum */
1058 	}
1059 	while (max_ctx < ia64_ctx.max_ctx) {
1060 		unsigned int old = ia64_ctx.max_ctx;
1061 		if (cmpxchg(&ia64_ctx.max_ctx, old, max_ctx) == old)
1062 			break;
1063 	}
1064 
1065 	if (ia64_pal_rse_info(&num_phys_stacked, NULL) != 0) {
1066 		printk(KERN_WARNING "cpu_init: PAL RSE info failed; assuming 96 physical "
1067 		       "stacked regs\n");
1068 		num_phys_stacked = 96;
1069 	}
1070 	/* size of physical stacked register partition plus 8 bytes: */
1071 	if (num_phys_stacked > max_num_phys_stacked) {
1072 		ia64_patch_phys_stack_reg(num_phys_stacked*8 + 8);
1073 		max_num_phys_stacked = num_phys_stacked;
1074 	}
1075 	platform_cpu_init();
1076 	pm_idle = default_idle;
1077 }
1078 
1079 void __init
1080 check_bugs (void)
1081 {
1082 	ia64_patch_mckinley_e9((unsigned long) __start___mckinley_e9_bundles,
1083 			       (unsigned long) __end___mckinley_e9_bundles);
1084 }
1085 
1086 static int __init run_dmi_scan(void)
1087 {
1088 	dmi_scan_machine();
1089 	return 0;
1090 }
1091 core_initcall(run_dmi_scan);
1092